[コラム] 画像生成の話 (Pixiv Fanbox)

こんにちは、mtosakです。

画像生成について、聞かれることが多かったので、コラム的な感じで書こうと思います。

（そんな大したことはやってないです）

用語の説明は省略するので、予備知識がないとよく分からない部分もあるかも。

前提

まず、私が使っているAIについてですが、Gemini Proの年間契約をしているので、Gemini系のAIを使うことが多いです。

ローカルの画像生成を利用可能な環境はあるのですが、私のゲーミングノートでは、ラップトップ用のGPUしか使えないので、最近はあまり使うことはありません。

スタンスとしては、既に頭の中にあるイメージを生成することを目標とします。どういう画像を生成するか、アイデアそのものは自分で決めて、ランダム性ができるだけ少なくなるように、細かくプロンプトを定義していくことが多いです。

（ランダム性を楽しむ流派もあると思うのですが、それとは違う作り方です）

作業の流れ

画像を生成するまでの流れは、以下のような感じです。

１．アイデアを出す

２．プロンプトを作る

３．画像生成する

４．プロンプトを修正する

（その後は、画像生成→プロンプトの修正の繰り返し）

近年は、AIで画像を生成して、「画像そのものを修正する」ような指示をすることもできます。

しかし、画像そのものを繰り返し修正してしまうと、様々な理由で画質が劣化してしまうことが多いので、「プロンプトそのものを修正する」ことで、一発で良い感じの画像を生成できるプロンプトの完成を目指します。

もちろん画像の差分を作る際など、参照画像そのものに対しての修正を加えることもあるのですが、「ここはこうじゃない」とか「ここはこうしてくれ」みたいな指示を重ねて、画像そのものを修正していくことはあまりしません。

プロンプトの生成

現在、画像生成用のプロンプトは、Antigravityというアプリケーションを使って生成することが多いです。（必ずしも、Antigravityを使う必要はないです）

Antigravityには、スキルという機能があって、LLMにやらせたいことを定義しておくことができます。そのため、そのスキルを使って、画像生成用の構造化されたプロンプトを作っています。

ちなみに、一般的なAIチャットの冒頭で、「以下のスキルを覚えてください」というような感じで指示をすると、疑似的に同等のことができるようになると思います。

私が実際に使っているスキルの定義は以下です。

（超適当に生成したものなので、いまいちなところがあるかも）

【スキル定義：ここから】

---

name: image_prompt_generator

description: Generates a highly detailed, structured text prompt for image generation models based on the user's concepts or requirements.

---

# Image Prompt Generator Skill

You are an expert prompt engineer specialized in generating highly detailed, structured text prompts for image generation models.

Your sole task is to construct and output a structured prompt block based on the user's input.

## Core Rules & Instructions

1. **Output Format**:

- You MUST output the constructed prompt wrapped inside a single Markdown code block (using triple backticks ```).

- Do NOT output any other text besides the code block. This means no greetings, no introductory or concluding sentences, no explanations, and no translations. Only the code block containing the prompt.

- Output the constructed prompt with a blank line between each property for readability.

- The prompt MUST start with the `Goal:` property and end with the `placement_of_objects:` property.

- Each property must be in the format: `[property_name]: [value]`.

2. **Prompt Composition**:

- Always write the prompt in **English**, translating and expanding upon any non-English concepts from the user.

- Fill in and extrapolate any missing details (such as specific outfits, hairstyles, lighting, color variations, background elements, and camera setups) to maximize output quality.

- If the user provides abstract character traits or background details not directly drawable (e.g., "enjoys weight training 3 times a week"), include them exactly as stated under a custom property like `character_background:` or `persona:` without modifying the text.

3. **Professional Prompting Strategies**:

- **Extreme Specificity**: Describe clothing, armor, objects, and textures in precise detail (e.g., instead of "fantasy armor", write "ornate elven plate armor etched with silver leaf patterns...").

- **Semantic Negatives**: Describe the desired scene positively to imply what should not be there (e.g., instead of "no cars", write "an empty, abandoned street with no signs of traffic").

- **Camera & Composition Control**: Use cinematic and photographic terms to control the shot (e.g., `wide-angle shot`, `macro shot`, `low-angle perspective`, `shallow depth of field`).

## Output Template & Example

When a user wants "A muscular Japanese male college athlete in his dorm room flexing double biceps," construct the prompt in this exact structured format:

Goal: A photograph of a muscular Japanese male college athlete in his dorm room

environment: Inside a college dorm room

space: A small, slightly messy dorm room with a sporty atmosphere

state: Flexing muscles, showing a shy smile

location: Standing in the center of the room

area_features: Wrestling posters, sports equipment, small desk

structure: Typical college dorm room layout

outfit: Red wrestling singlet

hairstyle: Short black hair, two-block cut with shaved sides and back, styled with hair wax

position: Standing, doing a double bicep pose

facial_expression: Shy, slightly embarrassed smile with a masculine, rugged look

body_type: Muscular, voluminous pro-wrestler-like physique

color: Vibrant summer colors

color_variation_1: Red (singlet)

color_variation_2: Tan skin

color_variation_3: Black hair

background_color_1: Off-white (walls)

background_color_2: Light wood (furniture)

lighting: Bright natural light from the window

weather: Hot summer day

season: Midsummer

time_of_day: Early afternoon

mood: Energetic, slightly embarrassed atmosphere

movement: Flexing muscles, shy head tilt

sound: Distant sounds of college activities

scent: Hint of sweat and masculine cologne

temperature: Hot and slightly stuffy

texture: Smooth, tanned skin, defined muscles

perspective: Slightly low angle, emphasizing muscular build

camera_settings: Close-up shot, wide aperture for shallow depth of field

height: Slightly below eye level

distance: Close-up (upper body to full body view)

size_of_objects: Athlete fills most of the frame

placement_of_objects: Athlete centered in the frame, dorm room visible in the background

【スキル定義：ここまで】

構造化のやり方自体は、SNSなどで数年前からよく見かける方法を踏襲しています。（同じような機能を持つプロンプトを数年前にDiscordでシェアしたこともあります）

近年の画像生成AIは、プロンプトの読解力が上がっているので、必ずしもプロンプトを構造化させる必要はないです。

ただ、構造化して、細かく画像の端から端まで定義しておくと、繰り返しプロンプトの修正箇所が特定しやすくなるのと、ランダム性を減らすことができて生成するたびに大きく変化してしまうということが少なくなると思います。

なお、使用するLLMの種類は、GeminiでいうとFlashレベルで良いのかなと思っています。性能が高いAIに作らせたほうがより安定しそうな気はしますが、試行回数が高くなると足りなくなるので、予算とコストを考えて選んだほうが良さそうです。

プロンプトを作ったあとは、実際に画像生成してみて、目視確認しつつ、プロンプトを修正するのを繰り返していく感じです。

繰り返して修正を重ねるとプロンプトは、非常に長くなってしまうことが多いです。長くなりすぎると、プロンプトの効きが悪くなるので、整理して縮小したほうが良い場合もあります。

ちなみに、画像生成する際には、Antigravityを使いません。

Antigravityで画像を生成することは可能ですが、少ない回数しか生成することができないので、あくまでも、Antigravityは画像生成用のプロンプトを生成するためだけに使っています。

プロンプトの例

繰り返し修正を重ねた結果でしかないので、プロンプトを見ても何の参考にならないと思うのですが、一応掲載しておきます。

# MCホモの品評会

【プロンプト：ここから】

Goal: An iPhone camera capture of four heavyweight Japanese male bodybuilders lined up on a stage, each with middle-aged men clinging to them in various bizarre positions with a hidden contest banner behind them.

environment: Stage of the contest hall at Kofu Citizen's Hall.

stage_backdrop: Matte navy blue velvet curtain, light-absorbing texture, measuring 6m high and 15m wide.

current_stage_sign: Hanging on the backdrop directly behind the bodybuilders is a large, horizontally long gold sign that is die-cut and silhouette-trimmed along the shape of the letters. The main gold text reads "MCホモの品評会". To the left and right of the text are gold pictograms of flexing muscular arms. Directly behind the main text is a large gold pictogram of a brain, which is surrounded by a circular gold laurel wreath and the circular gold text reading "ALL JAPAN CONTROLLED MUSCLE". This sign is positioned directly behind the bodybuilders' bodies; it is mostly hidden and obscured by their massive physiques, with only small fragments of the gold letters and emblem visible through the gaps between the athletes.

discarded_stage_sign: Leaned carelessly against the wall at the back right corner of the stage is the discarded original logo sign. It is a large, rectangular white plastic board with blue and black lettering that reads "ALL JAPAN CLASSIC MUSCLE". It has a realistic dark-green laurel wreath and a classic grey iron dumbbell emblem, with absolutely no gold or metallic colors to differentiate it from the new sign.

stage_floor: Deep dark-brown solid oak wood floor with a high-gloss finish, reflecting the warm overhead lights.

lighting_setup: Heavy overhead warm-toned halogen spotlights (3200K) casting deep dark shadows under the pectorals, abs, and thighs. Sharp cool-white LED rim lights (5600K) shining diagonally upwards from the rear stage floor, creating a bright outline around the bodybuilders' silhouettes.

camera_specifications: Captured on an iPhone. Wide lens, tight full-body framing. The top of the bodybuilders' heads touches the very top edge of the image frame, and the soles of their feet touch the very bottom edge of the image frame, leaving virtually no margins or empty space at the top and bottom. Crisp details, no artificial camera degradation.

interaction_rules: The contact is completely one-sided. The bodybuilders maintain professional, serious, deadpan facial expressions, totally ignoring the middle-aged men and pretending they are not there. The bodybuilders do not hug, hold, or react to the men; they only focus on holding their front double biceps pose with arms bent and both armpits fully exposed. All bodybuilders have smooth, hairless, completely shaved armpits for the competition.

bodybuilder_1_ohno_takuya:

name: Ohno Takuya

age: 28

physique: Height 174cm, weight 83kg. Thick arms, chest 130cm, waist 75cm, arm 50.5cm, thigh 67.5cm.

appearance: High fade haircut with swept-up top, short mustache and chin beard, dark tan.

outfit: Deep red posing trunks.

pose: Front double biceps pose with arms bent at 90 degrees, showing both of his smooth, shaved armpits. His face is serious and blank.

attached_men:

- intruder_4: A flabby 43-year-old Japanese man in boxer briefs, with a solid black censor bar covering his eyes. He is sticking his tongue out to lick the bodybuilder's cheek while sliding one hand directly inside the front of the bodybuilder's deep red posing trunks, groping his crotch.

bodybuilder_2_sugiyama_keisuke:

name: Sugiyama Keisuke

age: 32

physique: Height 182cm, weight 90kg. Tallest and heaviest, massive frame, chest 138cm, waist 81cm, arm 51.0cm, thigh 72.0cm.

appearance: Tight military style buzz cut, clean-shaven face, dark tan.

outfit: Purple metallic posing trunks.

pose: Front double biceps pose with arms bent at 90 degrees, showing both of his smooth, shaved armpits. His body remains facing completely forward, but his head is forced to turn to the side due to the intruder. His expression is serious.

attached_men:

- intruder_3: A stout, out-of-shape 46-year-old Japanese man in white briefs, with a solid black censor bar covering his eyes. He is standing in front of the bodybuilder, reaching up to grab the bodybuilder's face with both hands, forcing the bodybuilder's head to turn sideways toward him, and leaning in to press his lips against the bodybuilder's lips in a kiss.

bodybuilder_3_okada_tsuyoshi:

name: Okada Tsuyoshi

age: 34

physique: Height 178cm, weight 88kg. Massive bulk, chest 136cm, waist 78cm, arm 51.5cm, thigh 71.0cm.

appearance: Shaved bald head (skinhead), dense dark stubble beard, tanned skin.

outfit: Shiny black posing trunks with subtle glitter.

pose: Front double biceps pose with arms bent at 90 degrees, showing both of his smooth, shaved armpits. His face is serious and blank.

attached_men:

- intruder_1_1: A flabby, out-of-shape middle-aged Japanese man in dark boxer briefs with a solid black censor bar covering his eyes. He has his hands placed on the bodybuilder's muscular chest. He is pressing his face and tongue directly to a point exactly 10cm above the bodybuilder's left armpit, licking the upper arm area. His tongue is pressed flat against this spot, partially obscuring the area.

- intruder_1_2: A flabby, out-of-shape middle-aged Japanese man in loose trunks with a solid black censor bar covering his eyes. He has his hands placed on the bodybuilder's defined abdomen. He is pressing his face and tongue directly to a point exactly 10cm above the bodybuilder's right armpit, licking the upper arm area. His tongue is pressed flat against this spot, partially obscuring the area.

bodybuilder_4_sakamoto_ryuji:

name: Sakamoto Ryuji

age: 29

physique: Height 180cm, weight 87kg. Tall and broad-shouldered, chest 133cm, waist 77cm, arm 49.0cm, thigh 69.5cm.

appearance: Fade haircut with a messy top, clean-shaven face, dark tan.

outfit: Crimson red posing trunks.

pose: Front double biceps pose with arms bent at 90 degrees, showing both of his smooth, shaved armpits. His face is serious and blank. Because of the intruder, he is forced into a wide-open, bow-legged stance.

attached_men:

- intruder_2: A soft-bodied, out-of-shape 48-year-old Japanese man with a receding hairline in white briefs, with a solid black censor bar covering his eyes. He is crouching low between the bodybuilder's thighs, crawling under his crotch, popping his head out from under the groin, grabbing both of the bodybuilder's thighs with his hands to force the wide-open stance, and kissing the inner side of the bodybuilder's right thigh.

color: Strong contrast of warm halogen spotlights and cool white rim lights against the navy blue curtain.

season: Midsummer.

time_of_day: Afternoon.

mood: Bizarre and surreal exhibition, deadpan athletic focus contrasted with aggressive, unsolicited physical contact.

texture: Oiled, muscular, vascular tanned skin, metallic gold letters, soft pale skin of the out-of-shape intruders.

perspective: Eye-level wide shot from the judges' perspective. The framing is so tight vertically that the top of the bodybuilders' heads and the bottom of their feet touch the upper and lower edges of the frame respectively, with no extra margin.

placement_of_objects: The four bodybuilders are lined up horizontally in the center of the frame in the exact order from left to right: Ohno Takuya, Sugiyama Keisuke, Okada Tsuyoshi, and Sakamoto Ryuji, filling the frame from top to bottom. The gold "品評会" sign hangs directly behind them, mostly blocked from view by the bodybuilders' bodies, with only fragments of the lettering visible between them. The old rectangular "ALL JAPAN CLASSIC MUSCLE" sign is visible lying messily on the floor in the back right corner.

【プロンプト：ここまで】

備考：

これは、画像生成用ではなく別のプロンプトを考えたり、作業をしたりしているときに、イメージを見たくなって作ったプロンプトです。

プロンプト内に、各人物に細かく名前や髪型や体型が定義されているのですが、事前にボディビル大会の参加者全員の定義を作っていて、そこから人物を借用した形になります。その名残です。通常はそこまで細かく指定する必要はないと思います。

＊

# 妖怪パンツ脱がせおじさん

【プロンプト：ここから】

Goal: A side-by-side, three-panel split image comparing three different clothing and situational states of a muscular 38-year-old Japanese father on the street in front of his house, captured from a single fixed camera position inside a bush. The aspect ratio is 21:9.

composition:

layout: side-by-side three-panel split

left_panel: image_1

middle_panel: image_2

right_panel: image_3

common:

environment: On the street in front of a modern suburban house in Japan, with a utility pole (telephone pole) visible in the background.

space: The street directly in front of the father's house. A two-story modern Japanese house facade and a utility pole are in the background, creating a realistic Japanese residential street atmosphere.

area_features: Well-maintained suburban house facade, street, neat driveway, utility pole (telephone pole) standing nearby in the background.

structure: Two-story modern Japanese house and a utility pole.

father:

profession: Foreign-affiliated company sales representative

hobby: Bodybuilding

body_type: Extremely muscular, superhero-like physique, standing at 188 cm (approx. 6'2") and weighing 125 kg (approx. 275 lbs) with a body fat percentage of 8%, exceptionally wide shoulders, dramatic V-taper (upside-down triangle torso), massive biceps and chest, far exceeding average Japanese build.

facial_features: Monolid eyes, and a strong, well-defined jawline

accessories: A simple wedding ring on the ring finger of his left hand.

upper_outfit: A sporty, high-performance sleeveless athletic tank top (charcoal grey with subtle neon green side accent lines, made of stretchable dry-fit fabric).

footwear: A pair of sporty and modern thick-soled high-tech running shoes (black base with neon green accent lines that match the tank top).

color_variations:

skin: Tan

hair: Black

shorts: Black

woman:

description: A neighbor woman standing extremely close to the father in the immediate foreground, mostly cropped out of the left frame (barely visible, with only a small portion of her shoulder/back showing).

style: She is heavily out of focus (blurred, shallow depth of field), serving as a subtle framing edge on the far left to create a candid, secretly-taken atmosphere. Her appearance, position, and pose remain completely identical across all three panels.

man:

body_type: Middle-aged with a chubby, out-of-shape build and noticeably thinning hair on top (the exact same man in both the middle and right panels).

outfit: A slightly worn-out, casual polo shirt and casual half pants (shorts) (the exact same outfit in both panels).

background_color_1: Beige or light grey of the house exterior

background_color_2: Greenery of front yard plants

weather: Extremely hot, perfectly clear and cloudless day

season: Midsummer

time_of_day: Bright midday

temperature: Warm and pleasant

texture: Extremely defined skin texture, natural skin sheen, veins and muscle striations visible without post-processing

color_style: High color saturation and vivid colors to reflect the intense midsummer sun, while keeping the contrast natural and soft (no high-contrast).

perspective: A candid, secretly-taken shot from a single, fixed camera position hidden inside a bush. To emphasize the hidden, voyeuristic perspective, a few heavily blurred leaves and green foliage are slightly visible ONLY in the top-left and bottom-right corners of the frame in all three panels. The camera position, framing, composition, background facade, and the cropped, blurred woman on the left edge are completely identical across all three panels, locking their coordinates and positions. The father is standing very close to the woman, with his body facing front towards the camera, but his head is turned to the left, looking at her.

camera_settings: Automatic focus is on the father in the midground, with a shallow depth of field making the foreground elements (the cropped woman on the left edge, and the foliage in the corners) heavily blurred.

distance: Full-body shot, capturing the father from head to toe with his entire body, legs, and feet visible in the frame. The spatial position, standing pose, leg stance, and foot placement of the father are perfectly locked in place and remain completely identical across all three panels.

size_of_objects: The father is the primary subject, with his massive physique occupying 90% of the image frame, dominating the composition, with his face positioned in the exact same spot in all three panels.

sound: Ambient suburban quietness

scent: A complex, heavy mixture of intense body odors; featuring a persistent damp laundry smell (defined as "modori-namagawaki-shu"), a strong and sharp underarm odor, and a sweet yet sour ketotic body odor.

images:

image_1:

Goal: The muscular father talking to a neighbor woman in front of his house, standing very close to her. The woman is on the left edge, mostly cropped out of the frame and heavily out of focus. The father is standing naturally with his hands on his hips, with his body facing front towards the camera, but his head is turned to the left to make eye contact as they converse. The shot looks like a candid snapshot taken secretly from a bush. No other man is present in this panel.

state: The father is standing naturally, with his body facing front towards the camera, with his hands placed firmly on his hips, standing very close to her. His head is turned to the left, looking at her and making friendly eye contact. He is completely unaware of the camera. The neighbor woman is mostly cropped out on the left edge, blurred and facing away. No other characters are present.

location: The father is centered in the midground, with his body facing front towards the camera. The cropped, blurred woman is on the left edge.

father:

lower_outfit: A pair of black, breathable sports shorts, worn normally.

footwear: A pair of sporty and modern thick-soled high-tech running shoes (black base with neon green accent lines), worn normally on his feet.

position: Standing naturally with his hands on his hips, standing very close to the woman, with his body facing front towards the camera, but his head is turned to the left, making eye contact.

facial_expression: A friendly, natural smile as he speaks to the neighbor.

man:

presence: Only his face and a small portion of his upper body are visible, peeking out from behind the utility pole.

position: Lurking in the background behind the father, mostly hidden behind the utility pole (telephone pole) on the street, with only his head and a slight part of his upper torso visible as he peeks out to secretly watch the father.

facial_expression: Grinning with his tongue sticking out, with a solid horizontal black bar covering his eyes to censor his face.

mood: Natural, casual, and candid, with a secret snapshot atmosphere.

movement: Static, a relaxed conversation pose.

color_variations:

father_skin: Tan

father_hair: Black

father_tank_top: Charcoal grey with neon green side accent lines

father_shorts: Black

father_shoes: Black base with neon green accents

woman_hair: Dark brown

woman_clothing: Light beige blouse and long denim skirt

man_polo_shirt: Faded green

man_shorts: Khaki

placement_of_objects: Blurred foliage in the top-left and bottom-right corners. The cropped, blurred woman is on the left edge. The father is in the midground center, standing very close to and facing towards her. The house facade is in the background.

image_2:

Goal: The exact same scene, composition, and fixed camera view as the first panel, but a chubby, balding middle-aged man has appeared. The father has one hand placed behind his head and the other hand resting on his hip, exposing his underarm on the side of the raised arm. The man stands closely beside the father, lifting the hem of the father's tank top to expose his abs. The father is completely unaware that his hand is behind his head and his underarm is exposed, still looking at the very close woman and making eye contact with her, completely oblivious to the man's actions and the camera. His face position and expression are completely identical to the first panel.

state: The father stands in the exact same spot and angle as the first panel. He has one hand placed behind his head and the other hand resting on his hip, exposing his underarm on the side of the raised arm. The cropped, blurred woman is on the left edge, identical to the first panel. The chubby man stands beside the father, grabbing and lifting the hem of the father's tank top to expose about half of his defined abs, and is licking the father's exposed underarm/torso. The father remains focused on the very close woman, making eye contact, completely unaware of his pose or the man's actions. The man's eyes are covered by a solid black censor bar.

location: The father is centered in the midground. The man is closely beside him. The cropped, blurred woman is on the left edge.

father:

lower_outfit: A pair of black, breathable sports shorts, worn normally.

footwear: A pair of sporty and modern thick-soled high-tech running shoes (black base with neon green accent lines), worn normally on his feet.

position: One hand placed behind his head and the other hand resting on his hip, exposing his muscular underarm on the side of the raised arm. His body is facing front towards the camera, but his head is turned to the left, maintaining his focus on making eye contact with the very close woman (completely ignoring the man and the camera).

facial_expression: A friendly, natural smile (identical to image_1).

upper_outfit_interaction: The hem of the tank top is grabbed and lifted by the other man, exposing about half of the father's defined abs.

man:

position: Standing closely beside the father, hugging him from the side, with his face turned toward the father's exposed raised underarm/torso and his tongue pressed against the bare skin to lick it, while using one hand to grab and lift the hem of the father's tank top.

facial_expression: Grinning with his eyes closed in pleasure, with a solid horizontal black bar covering his eyes to censor his face.

mood: A bizarre contrast between the father's oblivious conversation and the intense, submissive actions of the man clinging to him.

movement: Static, the father holding the hand-behind-head and hand-on-hip pose while the man clings to his side.

color_variations:

father_skin: Tan with warm sheen

father_hair: Black

father_tank_top: Charcoal grey with neon green side accent lines

father_shorts: Black

father_shoes: Black base with neon green accents

woman_hair: Dark brown

woman_clothing: Light beige blouse and long denim skirt

man_skin: Fair with slight redness

man_polo_shirt: Faded green

man_shorts: Khaki

placement_of_objects: Blurred foliage in the top-left and bottom-right corners. The cropped, blurred woman is on the left edge. The father and the man are in the midground center, standing very close to her. The house facade is in the background.

image_3:

Goal: The exact same scene, composition, and fixed camera view as the first panel, but the father is now shirtless, having taken off his tank top which lies on the ground. He is posing in a front double biceps pose, wearing glittering metallic gold posing trunks. His black sports shorts have been pulled down to his ankles by the chubby man who is now crouching low behind his legs. Critically, the father is still looking towards the very close neighbor woman, making eye contact, completely unaware of the camera and the man at his feet. His face direction, gaze, head angle, and friendly expression are completely identical to the first and second panels. The woman remains cropped and blurred on the left edge, identical to the previous panels.

state: The father stands in the exact same spot and angle as the previous panels, but is now shirtless and posing with front double biceps. His tank top lies on the ground. He wears gold posing trunks, and his black sports shorts are pulled down to his ankles. The chubby man is crouching low behind the father's legs, holding the slipped-off shorts while looking directly at the camera lens with a creepy smirk and sticking his tongue out. The father's face direction, head angle, and friendly expression are completely identical to the previous panels, still looking at the very close neighbor woman and making eye contact with her, completely unaware of the camera and the man.

location: The father is centered in the midground. The man is crouching low directly behind the father's legs. The cropped, blurred woman is on the left edge.

father:

midsection_outfit: A pair of glittering, stage-ready metallic gold bodybuilding posing trunks.

discarded_outfit: The sleeveless athletic tank top, now lying taken off and discarded on the ground near his feet.

lower_outfit: The black sports shorts pulled all the way down to his ankles and resting on the ground.

footwear: A pair of sporty and modern thick-soled high-tech running shoes (black base with neon green accent lines), worn normally on his feet (with the slipped-off black sports shorts resting around his ankles just above the shoes).

position: Front double biceps pose, arms raised and flexing, standing very close to the very close neighbor woman, with his body facing front towards the camera, but his head is turned to the left, making eye contact, with his face direction, gaze, and head angle completely identical to image_1 and image_2 (completely ignoring the man and the camera).

facial_expression: A friendly, natural smile directed at the very close neighbor woman (completely identical to the expression and smile in image_1 and image_2).

color_variations:

father_skin: Golden-tan

father_hair: Black

father_posing_trunks: Glittering metallic gold

father_shorts: Black (slipped down to ankles)

father_shoes: Black base with neon green accents

woman_hair: Dark brown

woman_clothing: Light beige blouse and long denim skirt

man_skin: Fair with slight redness

man_polo_shirt: Faded green

man_shorts: Khaki

man:

position: Crouching low behind the father's legs, holding the slipped-off black sports shorts around the father's ankles, while tilting his head up to smirk and stick his tongue out directly at the camera lens.

facial_expression: Sly, grinning smirk directed at the camera with his tongue sticking out, with a solid horizontal black bar covering his eyes to censor his face.

mood: A dramatic, bizarre climax; the father flexes while maintaining a normal conversation with the neighbor woman, completely oblivious to the man at his feet and the hidden camera.

movement: Static, the father holding a powerful flexed pose while the man smirks.

constraints: Only one pair of black sports shorts must exist on the father's body/legs. The shorts must only exist slipped off at his ankles in this panel.

placement_of_objects: Blurred foliage in the top-left and bottom-right corners. The cropped, blurred woman is on the left edge. The father is in the midground center, standing very close to her, the second man is crouching low directly behind him at his feet, and the house facade serves as the background.

【プロンプト：ここまで】

備考：

ズボンだけ脱がせたらエッチかなという発想でした。

最終的に妥協して、ポージングトランクスを履かせました。